JARID1B enables transit between distinct states of the stem-like cell population in oral cancers

Abstract

The degree of heterogeneity among cancer stem cells (CSC) remains ill-defined and may hinder effective anti-CSC therapy. Evaluation of oral cancers for such heterogeneity identified two compartments within the CSC pool. One compartment was detected using a reporter for expression of the H3K4me3 demethylase JARID1B to isolate a JARID1B^high fraction of cells with stem cell-like function. JARID1Bhigh cells expressed oral CSC markers including CD44 and ALDH1 and showed increased PI3-kinase (PI3K) pathway activation. They were distinguished from a fraction in a G₀-like cell cycle state characterized by low reactive oxygen species and suppressed PI3K/AKT signaling. G₀-like cells lacked conventional CSC markers but were primed to acquire stem cell-like function by upregulating JARID1B, which directly mediated transition to a state expressing known oral CSC markers. The transition was regulated by PI3K signals acting upstream of JARID1B expression, resulting in PI3K inhibition depleting JARID1B^high cells but expanding the G₀-like subset. These findings define a novel developmental relationship between two cell phenotypes that may jointly contribute to CSC maintenance. Expansion of the G₀-like subset during targeted depletion of JARID1B^high cells implicates it as a candidate therapeutic target within the oral CSC pool.

INTRODUCTION

CSCs are defined by unlimited self-renewal and differentiation to states lacking growth potential. They have been isolated in oral squamous cell carcinomas (OSCCs) using multiple markers including CD44 and ALDH1 (1–3) and express regulators of embryonic stem cells including Oct4, Sox2, Nanog, and Bmi-1 (4–6). However, epigenetic plasticity among CSCs may prevent assignment of a single discrete molecular profile. Such plasticity among adult epithelial stem cells gives rise to quiescent and proliferative subsets that interconvert to support tissue homeostasis and repair (7, 8). Thus, similar heterogeneity among oral CSCs may drive cancer progression and shape therapy responses.

Although quiescence is not an essential stem cell trait, it is commonly attributed to adult stem cells and CSCs, where it may underlie generalized resistance to therapies targeting proliferating cells. As such, quiescence is defined by exit from the cell cycle while retaining cell division potential. One approach to defining quiescent cancer cells uses Pyronin-Y staining to gate cells with low total RNA within the G₀/G₁ fraction. Such cells exhibit the high p27^kip1 and low reactive oxygen species (ROS) of the G₀ state (9, 10). A related breast cancer subpopulation was isolated based on low ROS and exhibits quiescent features including suppression of proliferation markers and increased Hes1 (11), an inhibitor of senescence and differentiation (12). Not meeting strict criteria for complete G₀ exit, these “G₀-like” cells downregulate PI3K-Akt signaling through proteasome-mediated Akt degradation (11, 13). The role of PI3K-Akt activation in allowing certain adult stem cells to exit quiescence (14) suggests the pathway might also permit G₀-like cells to exert CSC function.

However, some molecular traits of G₀-like cells diverge from those described for CSCs, making their relationship unclear. Specifically, PI3K activation observed in putative CSCs of various cancers renders them sensitive to PI3K inhibitors (15–18). Furthermore, the ROS^low feature of G₀-like cells is incongruous with the high oxidative metabolism observed in melanoma CSCs that express increased levels of the H3K4me3 demethylase JARID1B (19). These JARID1B^high melanoma cells are slow-cycling (20) but are not known to express markers of quiescence. JARID1B, a transcriptional repressor, targets genes involved in diverse cellular processes including development, differentiation, and cell cycle (21–23). It is overexpressed in various cancers (23–25). In OSCC, JARID1B silencing is observed to attenuate sphere formation and stem cell marker expression (26). Additional findings have led to appreciation of JARID1B’s role in maintenance or differentiation of normal and malignant stem cell phenotypes (21, 27–29), suggesting broader utility as a CSC marker.

The unknown degree of identity between G₀-like and JARID1B^high cells led us to evaluate their function, overlap, and developmental relationship. Here we demonstrate that they are distinct subpopulations in OSCCs. Despite sharing stem cell-like function, only JARID1B^high cells displayed conventional CSC-associated markers, PI3K pathway activation, and PI3K inhibitor sensitivity. However, G₀-like cells were primed to enter a JARID1B^high state and required JARID1B to exert stem cell-like function. Our results have implications for maintenance of a dynamic CSC pool and provide insight into adaptation of oral CSCs to PI3K inhibition.

MATERIALS AND METHODS

Cell lines, clinical specimens, vectors

Cells were maintained in 1:1 DMEM/F12 with 400ng/ml hydrocortisone, 10% FBS, and 50μg/ml gentimycin. SCC9 cells were obtained from ATCC. LNT14 cells are described (30). VU147T and CAL33 cells were gifts from Dr. Hans Joenje, VU Medical Center, Netherlands and Dr. Jennifer Grandis, University of Pittsburgh, respectively. Cell lines were authenticated using Identify Mapping Kit (Coriell, Camden, NJ). Clinical specimens were obtained under University of Pennsylvania IRB protocol #417200 or Philadelphia VA protocol #01090. Lentiviral vectors pLU-JARID1Bprom-EGFP-Blast, pLU-CMVprom-EGFP-Blast, and pLKO-shJARID1B (20), transient expression vector pBIND-RBP2H1 (JARID1B) (31), and retroviral vectors pWZ-neo-myrAKTΔ4-129-ER/A2myrAKTΔ4-129-ER (32, 33) are described.

Flow cytometry (FC) and tumor cell purification

Antibodies are in Supplementary Table S1. FC was performed using AstriosEQ (Beckman-Coulter, Miami, FL) or LSRII (BD, San Jose, CA) instruments. Minced tumors were disrupted in 1mg/ml collagenase-IV using gentleMACS Dissociator (Miltenyi, San Diego, CA). Suspensions were agitated at 37°C for 1 hour before repeating disruption and 40μm filtering. Mouse stromal cells were removed from xenografts by pretreating with anti-mouse CD16/32-Fc-Block and positive selection of human cells using anti-HLA-ABC. Mouse cell contamination was monitored using anti-mouse-H-2K^d. For human tissues, tumor cells were purified by negative selection with anti-CD45/CD31 as described (34). Dead cells were excluded by 7-AAD throughout.

Detection of G₀-like cells, ROS, label retention, ALDH activity

Tumor cells at 10⁶ cells/ml were incubated with 4μM Hoechst-33342 (Life Technologies, Grand Island, NY) followed by 1μg/ml Pyronin-Y (Sigma-Aldrich, St. Louis, MO) at 37°C for 30 min. G₀-like cells were FACS-purified by setting PyroninY^low gates within the G₀/G₁ peak of the Hoechst-33342 profile. To detect ROS, cells were incubated with 2.5μM H₂DCFDA (Life Technologies) for 20 min at 37°C. PKH26/67 (Sigma-Aldrich) or CellTrace^™ Violet (Life Technologies) were used per manufacturer instructions. 100% labeling was confirmed by FC; labeled cells were analyzed after 10 days of culture. Aldefluor assays were performed per manufacturer instructions (Stemcell, Vancouver, Canada).

Western blotting

Protein lysates were made from equal cell numbers using Laemmli buffer, separated on 10% ECL gels (GE, Pittsburgh, PA), and transferred to nitrocellulose using the Trans-Blot® System (Bio-Rad, Hercules, CA). Antibodies (Supplementary Table S1) were incubated at 4°C overnight. After washing, blots were incubated with anti-Rabbit/Mouse IgG-DyLight and analyzed using a LI-COR Odyssey System (LI-COR, Lincoln, NE).

G₀-like cell detection in tissue

Paraffin-embedded samples were labeled using a multi-step tyramide-amplified protocol as described (35). Following antigen retrieval and serum-free protein block, each labeling cycle consisted of primary antibody (Supplementary Table S1), secondary antibody conjugated to horseradish peroxidase, and TSA conjugated to a fluorophore (FITC/CY3/CY5, Perkin Elmer, Inc). Images were acquired on a Nikon Ti confocal microscope (60X). Tumor cells were distinguished with pan-cytokeratin staining. G₀-like cells were identified based on the following pattern: DAPI+/AKT1^low/H3K9me2^low/HES1^high. Cells were counted in 10 random fields per section using Image J software. A blinded observer semi-quantitatively assessed ‘high’ or ‘low’ fluorescence relative to background from isotype controls. Signals were designated ‘high’ if the ratio of corrected total fluorescence was >2× compared to three ‘low’ cells in the same image.

Sphere formation

10 cells/well were cultured in ultralow attachment 96-well plates (Corning, Corning, NY) in serum-free complete MEGM (Lonza, Basel, Switzerland) for 14 days. Spheres were counted and imaged using a Leica DM IRB inverted microscope and iVision software (Biovision, Chester Springs, PA). Spheres were propagated by Accutase® (Life Technologies) dissociation for 20 minutes and trypan blue dead cell exclusion before re-plating.

Mouse experiments

Non-obese diabetic/severe combined immunodeficient/interleukin-2 receptor γ-chain-deficient (NSG) mice were utilized under Wistar Institute IACUC protocols 112652/112655. Tumors were generated by subcutaneous flank injection of cells in 100μl Matrigel (Corning). Tumors for drug studies were grown from 10⁶ cells to a volume of ~50mm³. GDC-0941 (SelleckChem, Houston, TX) in 0.5% methylcellulose/0.2% Tween80 was administered at 100mg/kg by daily oral gavage.

Real-time reverse transcription PCR and RNA-Seq

RNA was isolated using the Qiagen RNeasy kit (Qiagen, Valencia, PA). cDNA was generated from 1μg RNA using RNA-to-DNA (Life Technologies) and PCR purification (Qiagen) kits. Expression was quantified using Power SYBRgreen Master Mix and a Step-One Real-Time PCR System (Life Technologies). Primers are in Supplementary Table S2. RNA-Seq was performed on LNT14 cell fractions from three independent FACS isolations. Multiplexed Illumina libraries were prepared using Illumina stranded mRNA kits, pooled, and sequenced to 100bp from one end of the insert. 236,908,957 aligning reads against the human genome (hg19) were detected using RUM v2.0.4. 215,252,518 uniquely-aligning reads were used to quantify transcripts. Differential expression was defined using EdgeR with a generalized linear model to account for donor linkage and detect inter-treatment differences. Multidimensional scaling plots confirmed significant donor effects. Differentially-expressed genes had FDR>10%. GSEA was performed using the MSigDB C2 v5.0 curated gene sets (>3000) to compare JARID1B^high, G₀-like and bulk cells. A log2-fold change between paired conditions was used as a ranking variable.

Statistics and TIC-Frequency

Data are expressed as mean ± standard error. At least three independent replicates were performed per experiment. ANOVA or t-tests evaluated differences among means. Tukey’s procedure tested pairwise differences for significant ANOVAs. An F-statistic tested for equal variances. For unequal variances, Welch’s ANOVA or t-statistics were used. Mann-Whitney U tests were used for variables lacking normal distribution. Differences between cumulative distributions for latency times were defined using Kaplan-Meier and exact log-rank tests. Tumor initiating cell (TIC) frequencies were estimated using ELDA software (http://bioinf.wehi.edu.au/software/elda/) (36).

RESULTS

G₀-like OSCC cells exhibit stem cell-like function

G₀-like cells were identified by FC within multiple OSCC cell lines using Hoechst-33342 and Pyronin-Y to detect RNA^low cells within the G₀/G₁ cell cycle peak (Fig. 1A left, Supplementary Fig. S1A). H₂DCFDA staining confirmed the fraction to be ROS^low (Fig. 1A right, Supplementary Fig. S1B left), consistent with the H₂DCFDA-based definition of G₀-like breast cancer cells (11). G₀-like cells in cell lines showed high p27^Kip1 and Hes1, along with decreased total Akt protein (Fig. 1B, Supplementary Fig. S1B right). Continued cyclin D1 expression implied a lack of complete G₀ cell cycle exit, supporting the “G₀-like” designation. Applying the Pyronin-Y-based isolation strategy to tumor cells purified from clinical specimens and patient-derived xenografts (PDXs) also detected G₀-like cells with low ROS and Akt (Supplementary Fig. S1C–D). Similar cells were stained by established methodology in sections of 9 human tumors and 2 derivative PDXs. Cells with the Hes1^high/Akt^low/H3K9me2^low staining profile of ROS^low G₀-like breast cancer cells (11) were detectable as a minority fraction in most samples (Fig. 1C). G₀-like fractions from cell lines were tested in clonal sphere assays, where they showed enhanced formation of primary spheres and propagation as secondary spheres (Fig. 1D). They also formed tumors at higher incidence and shorter latency than the non-G₀-like fraction upon xenotransplantation at low cell dose (Fig. 1E left). G₀-like cells isolated from PDXs also had enhanced tumor-initiating capacity, forming tumors with similar efficiency to CD44^high cells (Fig. 1F top) despite being CD44^low (bottom, Supplementary Fig. S1E). Together, these results established that G₀-like OSCC cells can display CSC function. Clinico-pathologic traits for patient tumors, along with their derivative PDXs and cell lines, are in Supplementary Table S3.

Figure 1. G₀-like cells exhibit stem cell-like function.

A, FC plot illustrates G₀-like gate in representative LNT14 cell line (left). H₂DCFDA profiles compare G₀-like to non-G₀-like fractions in OSCC lines (right). *p<0.001,**p<0.0001. B, Western blots (WBs) of G₀-like and unfractionated (total) cells from OSCC lines. C, Confocal IF of patient tumor and derivative PDX stained for H3K9me2 (yellow), Hes1 (red), pan-Akt (green), and DAPI (blue). Arrows indicate G₀-like cells (H3K9me2^low/HES1^high/pan-AKT^low) (left, Bar=10μm). Staining quantifies G₀-like cells in 9 patient tumors (right). D, Primary and secondary sphere-formation by G₀-like vs. non-G₀-like fractions of OSCC lines. *p<0.05,**p<0.0001. E, Xenograft growth of G₀-like vs.non-G₀-like fractions from cell lines LNT14 (100 cells/mouse, n=6/group) and VU147T (1000 cells/mouse, n=6/group). *p<0.025,**p<0.0025. F, Xenograft growth (top) and cell surface CD44 (bottom) from G₀-like, CD44^high, and bulk fractions of OCTT2 PDX (2000 cells/mouse, n=4/group). Latency for G₀-like and CD44^high cells were shorter than for bulk cells (p<0.025).

JARID1B is a distinct basis for detecting CSC function

The quiescent and stem cell-like traits of G₀-like cells appeared comparable to those of melanoma cells expressing high JARID1B (20). To assess JARID1B’s role in the oral CSC pool, a promoter-based reporter for JARID1B transcription, J1BpromEGFP (20), was stably expressed in OSCC cell lines. LNT14 and SCC9 cells with 5% highest EGFP expressed increased JARID1B mRNA and protein (Fig. 2A, Supplementary Fig. 2A) and were designated JARID1B^high. All subsequent studies were standardized to assess changes in JARID1B^high population size relative to this 5% EGFP^high reference gate in control cells. The label retention and CSC function previously seen in a JARID1B^high state defined identically for melanoma (20) were evaluated in OSCC. Fluorescent membrane labelling showed that, in contrast to the G₀-like fraction, JARID1B^high OSCC cells were highly label-retaining over 10 days, indicating low-turnover (Supplementary Fig. S2B). However, JARID1B^high cells were distributed outside the G₀-like gate instead showing G₂M expansion (Fig. 2B, Supplementary Fig. S2C–D), suggesting slowed G₂M transit over G₀-like arrest. Decreased JARID1B reporter function and protein in the G₀-like gate (Supplementary Fig. S2E) distinguished the two states beyond cell cycle status, raising the possibility that JARID1B^high and G0-like cells are distinct subpopulations. Thus, subsequent studies standardized comparison of mutually exclusive G₀-like and JARID1B^high gates to a “bulk” gate that excludes both subsets (Supplementary Fig. S3A).

Figure 2. JARID1B is a distinct basis for detecting CSC function.

A, Standardized 5% JARID1B^high reference gate illustrated in LNT14_J1BpromEGFP cells (left). JARID1B expression in the EGFP^high fraction of LNT14_J1BpromEGFP or control LNT14_CMVpromEGFP cells by QRT-PCR (middle *p<0.025). JARID1B protein in JARID1B^high cells by WB (left, values indicate JARID1B normalized to loading control). B, Pyronin-Y/Hoechst-33342 plots distinguish cell cycle profiles of G₀-like and JARID1B^high fractions. *p<0.0005. C, Profiles show JARID1B^high fraction size in spheres of LNT14_J1BpromEGFP cells based on the 5% reference gate set in monolayer culture. *p<0.01. D, Primary and secondary sphere-formation by JARID1B^high vs. total cells. *p<0.0005,**p<0.0001. E, Xenograft formation by G₀-like, JARID1B^high, or bulk LNT14_J1BpromEGFP cells (100 cells/mouse, n=6/group). *p<0.05,**p<0.0025. F, Localization of JARID1B^high cellsto the top 10% CD44^high fraction in cell lines (left) was quantified (right). *p<0.05,**p<0.005. G, Localization of JARID1B^high PDX cells to the top 10% CD44^high fraction (left) and JARID1B expression by FC in this CD44^high fraction (right), determined by intracellular JARID1B IF after surface CD44 staining.

In functional analyses, JARID1B^high cells defined by the 5% reference gate in monolayer culture increased in clonal spheres (Fig. 2C, Supplementary Fig. S3B) and showed higher primary and secondary sphere-forming capacity (Fig. 2D). JARID1B^high cells from two lines formed xenografts with comparable efficiency to G₀-like cells but higher incidence and shorter latency than bulk cells, with all fractions producing tumors with indistinguishable histology (Fig. 2E, Supplementary Fig. S3C–D). Limiting dilution showed JARID1B^high LNT14 cells to form 100% tumors even at a dose of 10 cells and the G₀-like subset to have markedly higher TIC frequency than bulk cells (Supplementary Fig. S3E). Reliance on a reporter to purify JARID1B^high cells prevented their direct isolation from PDXs for functional testing. However, their overlap with the CD44^high fraction by was supported by FC analysis of cell lines and PDX tumors (Fig. 2F–G left, Supplementary Fig. S3F top) and detection of elevated JARID1B expression in CD44^high PDX cells (2G right, Supplementary Fig. S3F bottom). Together, these data suggested that JARID1B^high and G₀-like cells are subpopulations with distinct molecular characteristics but similar stem cell-like function.

JARID1B^high and G₀-like cells show divergent molecular profiles and PI3K signaling

Shared stem cell-like function between G₀-like and JARID1B^high cells despite divergent CD44 expression justified further molecular comparisons. Only JARID1B^high cells increased expression of the pluripotency-related factors Oct4 and Bmi1 (Fig. 3A) and another oral CSC marker ALDH1A1 (Fig. 3B left), the isoform underlying increased ALDH activity in OSCC (37). High ALDH activity in Aldefluor assays also corresponded with high levels of JARID1B (Fig 3B right). Broader comparison of G₀-like, JARID1B^high, and bulk cells was performed by mRNA sequencing. Pairwise comparison of G₀-like versus JARID1B^high subsets to bulk cells revealed 61 genes upregulated by both subsets (Fig. 3C, Supplementary Table S4–S7), including most transcripts increased in G₀-like cells. This significant overlap suggested a close relationship between the two states; however, minimal overlap existed between genes downregulated by G₀-like versus JARID1B^high cells (Supplementary Fig. S4A). Hierarchical clustering of gene-scaled expression levels confirmed that G₀-like and JARID1B^high cells clustered more closely to each other than to bulk cells (Supplementary Fig. S4B). Among the gene sets most significantly upregulated in G₀-like and JARID1B^high cells (Supplementary Tables S8–S10) were signatures defined in mammary and stromal stem cells (38, 39) (Fig 3D left, Supplementary Fig. S4C left), consistent with their shared CSC function. Notably, these gene set enrichment analysis (GSEA) profiles were more prominent in JARID1B^high over G₀-like cells (Fig. 3D right, Supplementary Fig. S4C right). JARID1B^high cells also more strongly upregulated an epithelial-to-mesenchymal transition profile from breast CSCs (40) (Fig 3E left). These findings reveal distinctions between G₀-like and JARID1B^high cells despite shared expression of a core gene set, with JARID1B^high cells consistently exhibiting a more conventional CSC-related profile.

Figure 3. JARID1B^high and G₀-like cells exhibit divergent molecular profiles and PI3K signaling.

A, WBs for indicated markers in G₀-like, JARID1B^high, and bulk LNT14 cells. B, QRT-PCR for ALDH1A1 in G₀-like, JARID1B^high, and bulk LNT14 cells (left) and JARID1B in Aldefluor+ vs. Aldefluor- cells. *p<0.05,**p<0.001. C, Genes upregulated relative to bulk in G₀-like and JARID1B^high LNT14 cells. D, GSEAs show the Lim_Mammary_Stem_Cell_Up gene set for the indicated pairwise comparisons. ES=enrichment score, FDR=false discovery rate. E, GSEAs for Sarrio_Epithelial_Mesenchymal_Transition_Up (left) and Reactome_Cell_Cycle (right) gene sets in G₀-like vs. JARID1B^high LNT14 cells. F, WBs of JARID1B^high vs. total cells.

Further analyses dissected the proliferative states of the subsets. Despite JARID1B^high cells retaining label, only G₀-like cells downregulated cell cycle-related gene transcription based on a Reactome database gene set (Fig. 3E right, Supplementary Fig. S4D). To further characterize this difference, proliferative oncogenic signals were compared across cell fractions. No differences were evident in MAPK or JAK/STAT signaling based on ERK and STAT3 phosphorylation, respectively (Supplementary Fig. S4E). The reduced PI3K activity anticipated in G₀-like cells based on low total AKT was confirmed at the level of phospho-AKT (Supplementary Fig. S2E, S4F). However, high phosphorylation of Akt and downstream target GSK3β indicated pathway hyper-activation in JARID1B^high cells relative to bulk (Fig. 3F, Supplemental Fig. S2E), adding to parallels between JARID1B^high cells and other CSC phenotypes (15–18). By contrast, AKT suppression in G₀-like cells suggested a distinct role in the CSC pool.

G₀-like cells are primed to become JARID1B^high

The functional properties shared by G₀-like and JARID1B^high cells prompted assessment of whether G₀-like CSC function arises from transition to a JARID1B^high state. The capacity of G₀-like versus bulk cells to upregulate JARID1B was tested by FACS-purification and re-culturing. G₀-like cells reconstituted the original JARID1B^high fraction by day 4, while bulk cells failed to do so fully over 14 days (Fig. 4A, Supplementary Fig. S5A). However, despite their higher capacity to become JARID1B^high, purified G₀-like cells proliferated comparably to bulk cells (Supplementary Fig. S5B). Thus, we tested whether G₀-like cells must return to rapid proliferation before becoming JARID1B^high or can transition directly. Comparison of cultures grown from purified G₀-like versus bulk cells labeled with CellTrace revealed two-fold more cells retaining label among JARID1B^high cells arising from the G₀-like subset (Fig. 4B). A proximate relationship between the G₀-like and JARID1B^high states was further supported by efficient G₀-like to JARID1B^high transition observed during sphere formation, where G₀-like cells generated spheres containing more JARID1B^high cells than bulk spheres (Fig. 4C). Here JARID1B upregulation by G₀-like cells was reflected by appearance of a second EGFP peak in a bimodal distribution, which was absent or less prominent in bulk-derived spheres (Fig. 4D). Together, these data support a developmental relationship in which G₀-like cells are primed to become JARID1B^high.

Figure 4. G₀-like cells are primed to become JARID1B^high.

A, Histograms illustrate JARID1B^high gate in purified G₀-like, bulk, and total LNT14_J1BpromEGFP cells on culture day 0, 4, and 7. Arrow highlights G₀-like cells shifting to the JARID1B^high fraction. Quantification below. *p<0.0025,**p<0.0001, n.s. not significant. B, CellTrace label-retention by JARID1B^high fraction arising from G₀-like vs. bulk cells after 10 days culture. *p<0.0025 C, JARID1B^high cell content in spheres generated from G₀-like vs. bulk LNT14_J1BpromEGFP cells. *p<0.0001. D, EGFP fluorescence within spheres from G₀-like, JARID1B^high, and bulk LNT14_J1BpromEGFP cells. Gates define the more intense fluorescence peak.

G₀-like cells exert CSC function by a JARID1B-dependent mechanism

The priming of G₀-like cells to become JARID1B^high led to testing whether JARID1B directly contributes to the transition. When JARID1B was reduced using an shRNA of established specificity (20) (Fig. 5A, Supplementary Fig. S5C), LNT14 cells in the G₀-like gate after JARID1B knockdown retained low ROS and AKT (Supplementary Fig. S5D). However, their sphere-forming capacity decreased to that of bulk cells, whose function was unaltered by shRNA (Fig. 5B, Supplementary Fig. S5E), and tumorigenesis by G₀-like cells diminished to the incidence and latency of bulk cells (Fig. 5C). These results could arise from reduced JARID1B preventing acquisition of oral CSC markers or from impairment of subsequent CSC function. The former was supported by accumulation of G₀-like cells upon JARID1B silencing (Fig. 5D, Supplementary Fig. S5F). Under silencing conditions, the 5% EGFP^high cells, which are expected to maintain high JARID1B transcription, were similar in frequency to control EGFP^high (i.e. JARID1B^high) cells but lacked high JARID1B mRNA (Supplementary Fig. S5G). Such JARID1B-silenced EGFP^high cells showed decreased sphere-formation (Fig. 5E left, Supplementary Fig. S5E) and lower Oct4 and Bmi1 (Fig. 5E right), supporting JARID1B’s direct role in acquisition of CSC markers and function. This finding agreed with marker and functional shifts observed during JARID1B knockdown or over-expression. Specifically, knockdown decreased surface CD44, ALDH1A1 mRNA, and ALDH activity (Fig. 5F, Supplementary Fig. S5H), while over-expression increased CD44 and ALDH1A1 coordinately with sphere-formation (Fig. 5G). In sum, these results indicate that JARID1B contributes to the function of G₀-like cells by driving them toward the JARID1B^high molecular profile.

Figure 5. G₀-like cells exert CSC function by a JARID1B-dependent mechanism.

A, JARID1B by QRT-PCR and WB (inset) in LNT14 cells expressing scramble (scr) or JARID1B (sh) shRNA. *p<5×10⁻⁵. B, Primary and secondary spheres from G₀-like and bulk LNT14 cells expressing sh vs. scr. *p<0.0005,**0.0001. C, Xenograft formation by G₀-like or bulk fractions from sh vs. scr LNT14 cells (100 cells/mouse, n=6/group). *p<0.025. D, G₀-like fraction size in two sh vs. scr cell lines. *p<0.025. E, Sphere formation (left) and Oct4/BMI1 WB (right) for EGFP^high and total fractions in sh or scr LNT14_J1BpromEGFP cells. Band densities are normalized to GAPDH. *p<0.0001. F, CD44 by FC and ALDH1A1 by QRT-PCR in sh vs. scr LNT14 cells. *p<0.0001. G, CD44 and ALDH1A1 in LNT14 cells after JARID1B cDNA or mock transfection (left). Sphere formation by transfected cells (right) *p<0.05,**p<0.025,***p<0.0005.

PI3-kinase signals regulate the G₀-like to JARID1B^high transition

Although JARID1B silencing blocked CSC marker acquisition, global pAKT levels were unaltered (Fig. 6A left) and AKT hyper-activation in EGFP^high cells was maintained (right). This result suggested that PI3K signals act proximal to JARID1B to drive the transition. To test this possibility, AKT was activated in LNT14_J1BpromEGFP cells carrying a myristoylated Akt variant fused to an estrogen receptor domain (myrAktER) that blocks activation in absence of 4-hydroxytamoxifen (4-OHT) (Supplementary Fig. S6A) (32). Addition of 4-OHT expanded JARID1B^high cells (Fig. 6B), supporting PI3K signaling driving entry into this state. Likewise, PI3K inhibition in vitro with LY294002 at a ~25% growth-inhibitory dose with minimal cytotoxicity (Supplementary Fig. S6B) decreased both the JARID1B^high fraction (Fig. 6C left, Supplementary Fig. S6C left) and JARID1B protein (Fig. 6C right). By contrast, the G₀-like fraction markedly expanded upon treatment with LY294002 or the clinical pan-PI3K inhibitor GDC-0941 (Fig. 6D, Supplementary Fig. S6C right) at doses producing modest growth inhibition (Supplementary Fig. S6B). G₀-like cells isolated after GDC-0941 treatment retained sphere-forming capacity (Fig. 6E left) and tumorigenicity (right), providing evidence of their PI3K inhibitor resistance. Xenografts established from LNT14_J1BpromEGFP cells were also treated in vivo with GDC-0941 at growth inhibitory doses (Fig. 6F left). Analysis of disaggregated residual tumors revealed no JARID1B^high cell enrichment and a trend toward their decreased percentage (Fig. 6F middle). Concurrently, there was greater than two-fold expansion of G₀-like cells (right). Expanded G₀-like subsets in treated tumors were verified by quantification of Hes1^high/Akt^low/H3K9me2^low cells in tumor sections (Fig. 6G). In sum, our findings support a model in which PI3K-dependent JARID1B upregulation drives the G₀-like to JARID1B^high state transition (Fig. 7). Therapeutic PI3K inhibition is therefore predicted to expand the G₀-like compartment while depleting JARID1B^high cells.

Figure 6. PI3-kinase signals regulate the G₀-like to JARID1B^high transition.

A, Akt and pAkt by WB in sh vs. scr LNT14 cells (left) and EGFP^high vs. total LNT14_J1BpromEGFP_shJARID1B cells (right). B, JARID1B^high fraction size in LNT14_J1BpromEGFP_myrAktER cells treated with 10nM 4-hydroxytamoxifen (4-OHT) for 72 hours. *p<0.0001. C, LNT14_J1BpromEGFP cells treated with 10μM LY294002 for 72 hours were analyzed for EGFP (left) and JARID1B by WB (right). *p<0.01. D, G₀-like cell content after LY294002 or 250 nM GDC-0941 treatment. *p<0.01,**p<0.0001 E, G₀-like and bulk LNT14_J1BpromEGFP cells purified after 72 hours GDC-0941 (GDC) treatment were analyzed for sphere (left) and xenograft (right, 100 cells/mouse, n=6/group) formation. *p<0.05. F, LNT14_J1BpromEGFP xenograft growth during GDC treatment (left, n=5/group). Tumors were analyzed for JARID1B^high cells (middle) and G₀-like cells (right). *p<0.025. G, Confocal IF of tumor sections from GDC- or vehicle-treated mice were stained as in Fig. 1E. Arrows indicate G₀-like cells (H3K9me2^low/HES1^high/pan-AKT^low) (left), which were quantified for GDC vs. vehicle groups (right, n=4 tumors/group). *p<0.05.

Figure 7. Model of G₀-like and JARID1B^high cells as related subsets of the oral CSC pool.

G₀-like cells exert stem cell-like function by PI3K-mediated entry to a JARID1B^high state. They are shown arising from rapid-cycling cells based on prior work (11). Dotted arrows represent transitions potentially impacting CSC homeostasis that are not addressed here.

DISCUSSION

A potential barrier to CSC-directed therapy arises from evidence that some cells lacking CSC markers retain stem cell-like functional capacity (20, 41, 42). Here we show that a G₀-like subset lacking standard oral CSC markers is primed for stem cell-like function through transition into a state expressing such markers. The transition was mediated by PI3K-dependent upregulation of the histone demethylase JARID1B. Regulation of a heterogeneous CSC pool by this mechanism may provide a novel basis for its homeostasis during therapy.

By establishing JARID1B’s function in mediating the CSC potential of G₀-like cells, our results expand understanding of its context-specific roles in malignant and normal stem cell homeostasis. JARID1B promotes either maintenance or differentiation of various normal stem cell populations (29, 43) at least partly through silencing lineage specification gene promoters (21, 44). In malignancy, increased JARID1B expression or amplification occur in multiple tumor types (23–25). Its oncogenic function is best characterized in breast cancer, where overexpression drives a luminal cell-specific expression program (23). By contrast, JARID1B is not overexpressed in melanomas but still underlies the function of a subset of JARID1B^high cells with a role in tumor maintenance and drug resistance (19, 20). Because JARID1B protein is widely detectable in OSCCs (26), CSC function associated with the rare JARID1B^high cells studied here may occur through dose-dependent effects on a subset of JARID1B-regulated genes. This gene regulation is also likely impacted by JARID1B’s known interactions with other transcriptional and epigenetic regulators that vary based on cellular context (25, 45–47).

Including G₀-like cells in the CSC pool in our model (Fig. 7) emphasizes their potential for stem cell-like function. G₀-like cells have been shown to arise from asymmetric division of rapidly-cycling cells as a stochastic event (11, 13). This event was mediated by a novel integrin-regulated signaling cascade leading to proteasomal Akt1 degradation via E3 ubiquitin ligase TTC3 (13) and may allow them to serve as an intermediate for transition to the JARID1B^high state. Although PI3K activation was observed here to drive this transition, it is unclear what signals initiate it or if additional pathways permit G₀-like cells to return to rapid proliferation without a JARID1B^high intermediate. Detailing such mechanisms may offer approaches to addressing G₀-like cells as a potential basis for innate therapy resistance.

The dichotomous PI3K activation in G₀-like versus JARID1B^high cells offers insight into the pathway’s role in homeostasis of the oral CSC pool and its adaptation to PI3K inhibition. Putative CSCs in multiple tumor types exhibit PI3K hyper-activation and small molecule inhibitor sensitivity (15–18). Our similar finding in JARID1B^high OSCC cells is particularly relevant given the frequent driving PI3K pathway alterations in OSCC (48), which include isolated PIK3CA mutations for some HPV-positive tumors (49). Our results suggest that G₀-like cells may be a reservoir that replenishes the JARID1B^high CSC state upon depletion by PI3K inhibition. Such effects may explain why PI3K pathway inhibitors generally produce modest clinical responses, despite the predicted sensitivity of the CSC pool. In this context, G₀-like cells may play a role in therapy resistance parallel to that of reserve stem cells recruited during injury responses in normal epithelia. For example, loss of proliferative Lgr5-positive intestinal stem cells through injury leads to their regeneration by quiescent Bmi1-positive cells that have limited roles in normal homeostasis (50). Similar plasticity among quiescent and proliferative adult stem cells is seen in multiple tissue types (7, 8) and likely is maintained and exploited by solid tumors.

Although absence of rapid proliferation in G₀-like and JARID1B^high cells may confer drug resistance to both subsets, their divergent PI3K activation and cell cycle states suggest a need for differing targeting strategies. One approach may be to block entry into the G₀-like state in combination with PI3K inhibition. In addition to addressing mechanisms regulating the G₀-like pool, further work is needed to test how molecular alterations in the PI3K pathway impact the size of G₀-like versus JARID1B^high subsets and their dynamics in response to PI3K inhibition. Likewise, studies are needed to test how the oral CSC pool is regulated by individual downstream components of the PI3K pathway. Together, such work may facilitate combination therapies to deplete all cell phenotypes cooperatively sustaining the oral CSC pool.